TL;DR

The Mastra job in the AI/ML testing pipeline skips every test that needs Automated Embedding or a live Atlas cluster. Turn those tests on, and add a weekly run against. 

Context

ai-ml-pipeline-testing runs each AI framework's MongoDB tests every day against current MongoDB versions. For Mastra, the test-mastra-js-local task runs the full @mastra/mongodb suite against local Docker containers.

The suite skips two groups of tests unless certain environment variables are set:

• Automated Embedding (MongoDB creates the embeddings itself): needs TEST_MONGODB_AUTOEMBEDDING=1 and a Voyage AI key. These can run on the local mongodb-atlas-local:preview image.
• Native Reranking (Atlas re-sorts results with a Voyage AI model): needs MONGODB_URI pointing to an Atlas cluster on MongoDB 9.0+ with Native Reranking enabled, and TEST_RERANKER=1. This can't run locally, because Atlas Local doesn't support $rerank.

The pipeline sets none of these, so both groups are skipped. The test-mastra-js-remote task is commented out.

Task

Goals: Any upstream or server change that breaks Automated Embedding or Native Reranking in Mastra shows up as a pipeline failure within a week.

Where: mastra-js/ and .evergreen/config.yml

Acceptance Criteria

• [ ] The Mastra pipeline runs the Automated Embedding tests, and they pass.
• [ ] The Mastra pipeline runs the Native Reranking tests against Atlas, and they pass.

Additional Context

The Atlas tests listed above read MONGODB_URI directly, so the remote task doesn't need to wait on NODE-7719 (making all vector tests accept a remote URI). That ticket can stay as a separate follow-up.